Download Visualization Viewpoints Data, Information and Knowledge in

Visualization Viewpoints
Editor: Theresa-Marie Rhyne
Data, Information and Knowledge in Visualization
In visualization, data, information and knowledge are
three terms used extensively, often in an interrelated context.
In many cases, they are used to indicate different levels of
Min Chen
abstraction, understanding or truthfulness. For example,
Swansea University
‘visualization is concerned with exploring data and
David Ebert
information [5]; ‘the primary objective in data visualization
Purdue University
is to gain insight into an information space’ [5]; and
‘information visualization’ is for ‘data mining and
Hans Hagen
knowledge discovery’ [4]. In other cases, these three terms
Technical University
are used to indicate data types, for instances, as adnominals
of Kaiserslautern
in noun phases, such as data visualization, information
visualization and knowledge visualization. These examples
Robert Laramee
suggest that data, information and knowledge could be both
Swansea University
the input and output of a visualization process, raising
Robert van Liere
questions about the exact role of data, information and
CWI, Amsterdam
knowledge in visualization.
Several attempts were made to clarify taxonomically the
Kwan-Liu Ma
terminology used in the visualization community (e.g.,
University of
[3,7,9]). However, the terms of data, information and
California, Davis
knowledge remain ambiguous. This article is not another
William Ribarsky
attempt to offer a different taxonomy for visualization.
University of North
Instead, we present a clarification that differentiates these
Carolina, Charlotte
three terms from the perspective of visualization processes.
Furthermore, we examine the current and future role of
Gerik Scheuermann
information and knowledge in the development of the
University of Leipzig
visualization technology.
Deborah Silver
Definitions of data, information and knowledge
Rutgers University
Since we can read data, grasp information and acquire
knowledge, we must first differentiate these three terms in
the perceptual and cognitive space. Because we can also
store data, information and knowledge in the computer, we
thereby must also differentiate them in the computational
space.
Perceptual and Cognitive Space
The Data-Information-Knowledge-Wisdom (DIKW)
hierarchy [1] is a popular model for classifying the human’s
understanding in the perceptual and cognitive space. The
origin of this hierarchy can be traced to the poet T.S. Eliot
[3]. Let  be the set of all possible explicit and implicit
human memory. The former encompasses the memory of
events, facts and concepts, and the understanding of their
meanings, context and associations. The latter encompasses
all non-conscious forms of memory, such as emotional
responses, skills, habits and so on [8]. We can thus focus on
three subsets of memory, data Õ , info Õ , and know Õ ,
where data, info, and know are the sets of all possible
explicit and implicit memory about data, information, and
knowledge, respectively.
There are many competing definitions of data,
information and knowledge, among which the definitions by
Ackoff [1] are frequently cited (see Table 1). Nevertheless,
there is a general consensus that data is not information, and
information is not knowledge. Without diverting from the
scope of this article, here we simply assume that data, info,
know are not mutually disjoint, and none of them is a subset
of another. Without losing generality, we can generalize
Table 1. Ackoff’s definitions of data, information and
knowledge in the perceptual and cognitive space [1].
Category
data
information
knowledge
Definition
symbols
data that are processed to be useful, providing answers
to ‘who’, ‘what’, ‘where’, and ‘when’ questions
application of data and information, providing answers to
‘how’ questions
Table 2. Our definitions of data, information and
knowledge in the computational space.
Category
data
Definition
computerized representations of models and attributes
of real or simulated entities
„ information data that represents the results of a computational
process, such as statistical analysis
„ knowledge data that represents the results of a cognitive process,
such as perception, learning, association, and reasoning
know to include also wisdom, and any other high-level of
understanding, in the context of DIKW hierarchy.
Computational Space
Let  be the set of all possible representations in
computer memory. Similarly, we may consider three subsets
of representations, data, info, and know. However, data is
an overloaded term in computing. For example, it is
common to treat programs as a special class of data. In
many cases, it is not possible to distinguish programs from
other data. Applying the same analogy, a computer
representation of a piece of information or knowledge is just
a particular form of data. A computer representation of
visualization is also a form of visual data.
We hence propose to use the definitions in Table 2 for
the following discussions. With such definitions, we have
data = , info Õ data, and know Õ data. The definitions in
Table 2 can easily be extended to include categories of raw
data, software, imagery data, mathematical models, and so
forth. This also makes sense when using the category names
as the adnominals in noun phases, such as software
visualization and knowledge visualization.
Figure 1 shows a typical visualization process, where
instances of data, information and knowledge in both
computational space and perceptual and cognitive space are
illustrated. Hence the purpose of visualization can be
rationalized by the difficulties for humans to acquire a
sufficient amount of information (Pinfo Õ info) or knowledge
(Pknow Õ know) directly from a dataset (Cdata Õ data). The
process of visualization is a function that maps from data to
the set of all imagery data, image. It transforms a dataset
Cdata to a visual representation Cimage, which facilitates a
more efficient and effective cognitive process for acquiring
Pinfo and Pknow.
A visualization process is a search process
Given a dataset Cdata, a user first makes some decisions
about visualization tools to be used for exploring the dataset.
The user then experiments with different controls, such as
styles, layout, viewing position, color maps, transfer
functions, etc. until a collection of satisfactory visualization
results, Cimage, is obtained. Depending on the visualization
tasks, satisfaction can be in many forms. For example, the
user may have obtained sufficient information or knowledge
about the dataset, or may have obtained the most
Interaction
computational space
Pknow
Cctrl
Visualization
Pinfo
Cimage
Cdata
perceptual and
cognitive space
Figure 1. A typical visualization process.
Interaction
computational space
Cctrl
Pknow
Visualization
Cimage
Pinfo
Cdata
Processing
Visualization
Cinfo
Cimage
perceptual and
cognitive space
supporting visualization pipeline
Figure 2. Information-assisted visualization.
Interaction
computational space
Cctrl
Pknow
Visualization
Cimage
Pinfo
Cdata
perceptual and
cognitive space
Processing
Reasoning
Cinfo
Cknow
Figure 3. Knowledge-assisted visualization with acquired knowledge
representations.
Interaction
computational space
Pknow
Visualization
Cimage
Pinfo
Cdata
perceptual and
cognitive space
Processing
Cdata
Cinfo
Processing
Reasoning
Cinfo
Information-assisted visualization
In recent years, an assortment of techniques were
introduced for visualizing complex features in data by
relying on information abstracted from the data. Note that
here we consider info in the computational space as well as
info in the perceptual and cognitive space. Figure 2
illustrates an information-assisted visualization process.
There are techniques that make use of information captured
in the visualization process to improve the efficiency and
effectiveness of visualization. Examples of such information
are given in Table 3.
Table 3 Examples of information used in visualization.
information categories
information about the input dataset
„ abstract geometric and
temporal characteristics
„ topological properties
examples
skeletons, features, events
contour tree for volume data,
vector field topology, tracking
graph for time-varying data
histogram, correlation, importance,
„ statistical indicators and
certainty, entropy, mutual informainformation measurements
tion, local statistical complexity
information about the results
color histogram, level of cluttering
information about the process
interaction patterns, provenance
Information about users’ perception response time, accuracy
knowledge-based system
Cctrl
appropriate illustration about the data to assist the
knowledge acquisition process of others.
Such a visualization process is fundamentally the same
as a typical search process, except that it is usually much
more complex than trying out a few keywords with a search
engine. In visualization, the tools for the ‘search’ tasks are
usually application-specific (e.g., network, flow, volume
visualization). The parameter space for the ‘search’ is
normally huge (e.g., exploring many viewing positions or
trying out many different transfer functions). The user
interaction for the ‘search’ sometimes can be very slow,
especially in handling very large datasets. This is depicted
in Figure 1 by a large interaction box that connects from the
user to the control parameters, Cctrl, which are also data.
In fact, over the past two decades, much of the emphasis
has been placed on improving the speed of visualization
tools, so the user can carry out the interactive ‘search’ faster,
can explore bigger parameter space, and hopefully find
satisfactory results quicker.
However, with the growing amount of data and
increasing availability of different visualization techniques,
the ‘search’ space for a visualization process is also getting
larger and larger. Like the internet search problem,
interactive visualization alone is no longer adequate.
Cknow
Reasoning
knowledge supporting infrastructure
other visualization processes
Figure 4. Knowledge-assisted visualization with simulated cognitive processing.
In information-assisted visualization, the user is
provided with a second visualization pipeline (see Figure 2),
which typically displays the information about the input
dataset, but can also present attributes of the visualization
process, the properties of the results, or characteristics of the
user’s perceptual behaviors. The user uses such information
to reduce the ‘search’ space for optimal control parameters,
hence making the interaction much more cost-effective.
Such techniques provide an intrinsic interface between
the scientific visualization and information visualization
communities. With the increasing size and complexity of
data, the use of information to aid visualization will
inevitably become a necessity rather than an option.
Knowledge-assisted visualization
In a visualization process, the knowledge of the user is
an indispensable part of visualization. For instance, the user
may assign specific colors to different objects in
visualization according to certain domain knowledge. The
user may choose certain viewing positions because the
visualization results can reveal more meaningful
information or a more problematic scenario that requires
further investigation.
Meanwhile, the lack of certain knowledge by the user is
often a major obstacle in deploying visualization techniques.
The user may not have received adequate training about
how to specify transfer functions. The user may not have
sufficient time or navigation skills to explore all possible
viewing positions.
Both scenarios suggest the need for knowledge-assisted
visualization. The objectives of knowledge-assisted
visualization include sharing domain knowledge among
different users, and reducing the burden upon users to
acquire knowledge about complex visualization techniques.
It also enables the visualization community to learn and
model the best practice, and to develop powerful
visualization infrastructures evolutionarily.
In fact, some general or domain knowledge has already
been incorporated into various visualization systems,
intentionally or unintentionally. For example, a default
transfer function in a volume visualization system may
capture the domain knowledge about a specific modality. If
one could collect a large repository of such knowledge, it
would be possible for a visualization system to choose an
appropriate transfer function according to the information
about the input datasets. Figure 3 shows a visualization
pipeline supported by a knowledge base, which stores the
knowledge representations captured from expert users.
Rule-based reasoning can be utilized to establish an
appropriate set, or several optional sets, of control
parameters, which can significantly reduce the ‘search’
space, especially for inexperienced users.
The shortcomings of such a system include the
difficulties in specifying comprehensively what knowledge
to capture and the inconvenience in collecting knowledge
from experts. This constrains the deployment of such a
system to specific application domains.
An alternative approach is to establish a visualization
infrastructure, where data about visualization processes are
systematically collected, processed and analyzed. Using
case-based reasoning, knowledge can be inferred from
cases of successes and failures, the common associations
between datasets and control parameters, and many other
patterns exhibited by the systems, the users and the
interactions. Such knowledge may include popular approach,
commonly-used parameter sets, best practice, optimization
strategy, and so forth. Figure 4 shows such an infrastructure.
Such an infrastructure is general-purpose, and can
support multiple application domains. It can potentially
enable applications to benefit from the best practice and
software developed for other applications. The development
of such an infrastructure can be built upon the advances in
other areas of computing technologies, including semantic
computing, autonomic computing, knowledge-based
systems, data warehousing, machine learning, and search
engine optimization.
Conclusions
Similar to the development of many other computing
technologies, for example, speech processing, computer
vision, web technology, one likely development path for
visualization is
„
„
„
„
from offline visualization
to interactive visualization,
to information-assisted visualization,
then to knowledge-assisted visualization.
Interactive visualization has reached a matured status.
There is a significant amount of ongoing development
currently in information-assisted visualization. With a large
amount of information collected locally and globally, it is
inevitable that there will be a transition to knowledgeassisted visualization.
As a discipline, visualization has thrived on helping
application users to transfer data (data) in the computational
space to information (info) and knowledge (know) in the
perceptual and cognitive space. As a discipline, we need
infrastructures to collect our own data about visualization
processes, and to transfer such data to information and
knowledge, which helps further our understanding as well as
enhance the visualization technology.
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
R. L. Ackoff, “From data to wisdom”, Journal of Applies
Systems Analysis, vol. 16, 1989, pp. 3-9.
S. K. Card, J. D. Mackinlay and B. Shneiderman, Readings in
Information Visualization: Using Vision to Think, Morgan
Kaufmann Publishers, San Francisco, 1999.
E. H. Chi, “A taxonomy of visualization techniques using the
data state reference model”, Proc. IEEE Symposium on
Information Visualization, 2000, pp. 69-75.
U. Fayyad, G. G. Grinstein and A. Wierse, Information
Visualization in Data Mining and Knowledge Discovery,
Morgan Kaufmann Publishers, San Francisco, 2002.
G. Scott, G. Domik, T.-M. Rhyne, K. W. Brodlie and B. S.
Santos, Definitions and Rationale for Visualization, http://www.
siggraph.org/education/materials/HyperVis/visgoals/visgoal2.htm,
(accessed in April 2008).
N. Sharma, The Origin of the “Data Information Knowledge
Wisdom” Hierarchy, http://www-personal.si.umich.edu/~nsharma/
dikw_origin.htm, (accessed in April 2008).
B. Shneiderman, “The eyes have it: a task by data type
taxonomy for information visualizations”, Proc. IEEE
Symposium on Visual Languages, 1996, pp. 336-343.
E. E. Smith and S. M. Kosslyn, Cognitive Psychology: Mind
and Brain, Pearson Prentice-Hall, Upper Saddle River, New
Jersey, 2007.
M. Tory and T. Moller, “Rethinking visualization: A highLevel taxonomy”, Proc. IEEE Symposium on Information
Visualization, 2004, pp.151-158.
Examples
There are many examples of information-assisted
visualization. On the other hand, the development of
knowledge-assisted visualization is very much in its
infancy. Here we selectively describe several examples of
information-assisted visualization in the literature, whilst
accentuating the use, or potential use, of knowledge in a few
visualization systems. These examples are intended to
reinforce the viewpoints of this article, rather than to
provide a comprehensive survey.
Examples of information-assisted visualization
Curve-skeleton
Curve-skeletons are 1D geometrical representations
abstracted from 3D objects in an input dataset. Such
information can be used to aid visualization tasks, including
virtual navigation, reduced-model formulation, visualization
improvement, and animation. For example, in virtual
endoscopy, curve-skeletons are used to specify collision
free paths for navigation through human organs [2].
Pre-determined ranking
In [6], a noticeable amount of generic knowledge is
captured as ranks of different visualization designs. This
enables the visualization system to automatically take users
through a design process for creating a visualization. The
stored ranks and ranking conditions are essentially a
collection of expert knowledge.
Figure 5. The local statistical complexity (LSC) of a flow around a delta wing
(gray triangle). Four streamsurfaces indicate the vortices on top of the delta
wing. The two isosurfaces in blue and light blue separate regions that hold LSC
values within the range [14.7;15] and [11;15] respectively. High LSC values point
the user to distinctive regions that may feature significant temporal events. The
image is rendered by Heike Jänicke, University of Leipzig [5].
Isosurface topology
Isosurface topology, which is typically represented as a contour tree,
provides an abstract insight into the structural relationship and connectivity
between isosurfaces in a dataset. In volume visualization, such information can
assist users in distinguishing features in different topological zones,
comprehending complex relationships between isosurfaces, and designing
effective transfer functions [8].
Local statistical complexity
Local statistical complexity (LSC) is an information-theoretic measure,
which tells how much information from the local past is required to predict the
dynamics in the local future. Given a time-varying dataset, we can assign each
data point an LSC value. Higher LSC values indicate regions that feature an
extraordinary temporal evolution, whereas, lower values indicate temporal
patterns that occur frequently in the dataset [5]. As demonstrated in Figure 5,
such information can assist users in generating a visualization that highlights
temporally-important features.
Data abstraction quality
Measuring the quality of visualization results, such as visual density and
clutter, provides users with useful guidance in synthesizing the most effective
visualization. One of such measurements is data abstraction quality, measuring
the degree to which the visualization results convey the original dataset. Such
information enables users to determine the optimal abstraction level for a given
visualization task, and to compare different visualization methods in terms of
their capability of maintaining dominant characteristics of the original dataset
while reducing the size and detail of the data [3].
Ontology mapping
The determination of visualization designs and
parameters should depend on the input data. One approach
is to extract semantic information from the input data, and
try to find the best match with the semantic information of
visualization designs (e.g., treemaps, graphs) and the
associated parameters (e.g., size, axes). In [4], three
ontologies, which are knowledge representations, are used
to store (a) the domain-specific semantics about a class of
input data, (b) the semantics about available visualization
designs, and (c) the ontological mapping from (a) to (b).
With these three ontologies, different visualization designs
are dynamically ranked according to the input data, and a
set of highly-ranked visualization designs are presented to
the user automatically.
Workflow management
VisTrails is a visualization infrastructure that provides
users with workflow management [1]. It is capable of
capturing and storing a huge amount of data about input
datasets, user interaction and visualization results in
visualization processes. VisTrails exhibits some of the
primary characteristics of the knowledge supporting
infrastructure shown in Figure 4, though it currently has
limited automated reasoning capability. Such an
infrastructure has great potential to be developed into an
infrastructure for knowledge-assisted visualization.
References
1.
2.
3.
4.
5.
Examples of knowledge-assisted visualization
Viewpoint mutual information
From Figures 2 and 3, we can observe that one transition path of
information-assisted visualization to knowledge-assisted visualization is to
automate the process of reasoning about the information abstracted from the
input data. A classical example of such a transition is [7], where viewpoint
mutual information (VMI) that measures the dependence or correlation between
a set of viewpoints and a set of objects in a dataset is used to determine the
optimal viewpoint. The fundamental difference between this approach and the
above-mentioned examples is that users do not make decision according to the
processed VMI. Instead, a relatively simple rule for minimizing VMI is used to
determine viewpoint transformation automatically. Such a rule can be seen as a
piece of knowledge hard-coded in the system.
6.
7.
8.
L. Bavoil, S. P. Callahan, P. J. Crossno, J. Freire, C. E.
Scheidegger, C. T. Silva and H. T. Vo, “VisTrails: enabling
interactive multiple-view visualizations”, Proc. IEEE
Visualization, 2005, pp. 135-142.
N. D. Cornea, D. Silver, P. Min, “Curve-skeleton properties,
applications, and algorithms”, IEEE Transactions on
Visualization and Computer Graphics, vol. 13, no. 3, 2007, pp.
530-548.
Q. Cui; M. Ward, E. A. Rundensteiner and J. Yang,
“Measuring data abstraction quality in multiresolution
visualizations”, IEEE Transactions on Visualization and
Computer Graphics, vol. 12, no. 5, 2006, pp. 709-716,
O. Gilson, N. Silva, P.W. Grant and M. Chen, “From web data
to visualization via ontology mapping”, Computer Graphics
Forum (special issue for EuroVis2008), vol. 27, no. 3, 2008, to
appear.
H. Jänicke, A. Wiebel, G. Scheuermann and W. Kollmann,
“Multifield visualization using local statistical complexity”,
IEEE Transactions on Visualization and Computer Graphics,
vol. 13, no. 6, 2007, pp. 1384-1391.
J. D. Mackinlay, P. Hanrahan and C. Stolte, “Show Me:
automatic presentation for visual analysis”, IEEE Transactions
on Visualization and Computer Graphics, vol. 13, no. 6, 2007,
pp. 1137-1144.
I. Viola, M. Feixas, M. Sbert and M. E. Gröller, “Importancedriven focus of attention”, IEEE Transactions on Visualization
and Computer Graphics, vol. 12, no. 5, 2006, pp. 933-940.
G. H. Weber, S. E. Dillard, H. Carr, V. Pascucci, B. Hamann,
“Topology-controlled volume rendering”, IEEE Transactions
on Visualization and Computer Graphics, vol. 13, no. 2, 2007,
pp. 330 - 341.